Skip to content
Agentic Engineering Primer

The Five Levels of AI Coding

This model adapts Dan Shapiro’s January 2026 framework. It is a simple way to judge how much of the coding loop the AI is carrying.

The Levels

Section titled “The Levels”
LevelNameDescriptionYour Role
0Spicy AutocompleteCopy-paste from ChatGPT, basic tab completionWrite most code yourself
1The Coding InternAI writes boilerplate and unimportant snippetsFull review of every line
2The Junior DeveloperPair programming with AI, real-time collaborationReview every line, guide direction
3The DeveloperMost code is AI-generated, you’re a full-time reviewerCode review, architecture decisions
4The Engineering TeamYou’re the PM or manager. Specs and plans drive the work.Define specs, review outcomes
5The Dark FactoryNo human code review. Only system verification.Design verification systems

Level Details

Section titled “Level Details”

Level 0: Spicy Autocomplete

Section titled “Level 0: Spicy Autocomplete”
  • Copy code snippets from ChatGPT
  • Use basic tab completion
  • Manually integrate everything
  • Prerequisites: None

Level 1: The Coding Intern

Section titled “Level 1: The Coding Intern”
  • AI writes boilerplate, tests, repetitive code
  • You review every line
  • You make all decisions
  • Prerequisites: Basic prompting skills

Level 2: The Junior Developer

Section titled “Level 2: The Junior Developer”
  • Real-time pair programming
  • AI suggests, you guide
  • Interactive back-and-forth
  • Prerequisites: Good prompts, context files

Level 3: The Developer

Section titled “Level 3: The Developer”
  • Most code is AI-generated
  • You’re a full-time code reviewer
  • You make architecture decisions
  • Prerequisites: Strong verification skills, TDD mindset

Level 4: The Engineering Team

Section titled “Level 4: The Engineering Team”
  • You write specs, AI writes code
  • Multiple agents working in parallel
  • You’re the PM/tech lead
  • Prerequisites: Robust test suites, CI/CD, monitoring

Level 5: The Dark Factory

Section titled “Level 5: The Dark Factory”

“Nobody reviews AI-produced code, ever. They don’t even look at it. The goal of the system is to prove that the system works.” — Simon Willison

  • No human code review
  • Only system verification
  • Trusting tests, not code
  • Prerequisites: Comprehensive conformance suites, 20+ years experience to design verification

Where to Start

Section titled “Where to Start”

If you’re new to AI coding: start at Level 1 or 2. Use AI for tasks you already know how to solve. That is how you learn when it is right and when it is not.

If you’re experienced: most professionals still live around Level 2 or 3. Level 4 and beyond needs real infrastructure.

Most readers should aim for Level 3 with strong verification practices.

Level 5 Warning

Section titled “Level 5 Warning”

Level 5 is real but rare. It requires:

  • Comprehensive conformance test suites
  • Continuous verification systems
  • Deep engineering experience to design the verification
  • Acceptance that you’re trusting the tests, not the code

Don’t aim for Level 5 until you’ve mastered Levels 2-4.

Progression Path

Section titled “Progression Path”
Level 1 → Level 2
  └── Learn prompting, add context files


Level 2 → Level 3
  └── Add TDD, strong verification, trust your review skills


Level 3 → Level 4
  └── Build robust CI/CD, test suites, learn spec-driven development


Level 4 → Level 5
  └── Build conformance suites, design verification systems

Next Steps

Section titled “Next Steps”